Rethinking earthquake prediction

--We re-examine and summarize what is now possible in predicting earthquakes, what might be accomplished (and hence might be possible in the next few decades) and what types of predictions appear to be inherently impossible based on our understanding of earthquakes as complex phenomena. We take predictions to involve a variety of time scales from seconds to a few decades. Earthquake warnings and their possible societal uses differ for those time scales. Earthquake prediction should not be equated solely with short-term prediction--those with time scales of hours to weeks--nor should it be assumed that only short-term warnings either are or might be useful to society. A variety of "consumers" or stakeholders are likely to take different mitigation measures in response to each type of prediction. A series of recent articles in scientific literature and the media claim that earthquakes cannot be predicted and that exceedingly high accuracy is needed for predictions to be of societal value. We dispute a number of their key assumptions and conclusions, including their claim that earthquakes represent a self-organized critical (SOC) phenomenon, implying a system maintained on the edge of chaotic behavior at all times. We think this is correct but only in an uninteresting way, that is on global or continental scales. The stresses in the regions surrounding the rupture zones of individual large earthquakes are reduced below a SOC state at the times of those events and remain so for long periods. As stresses are slowly re-established by tectonic loading, a region approaches a SOC state during the last part of the cycle of large earthquakes. The presence of that state can be regarded as a long-term precursor rather than as an impediment to prediction. We examine other natural processes such as volcanic eruptions, severe storms and climate change that, like earthquakes, are also examples of complex processes, each with its own predictable, possibly predictable and inherently unpredictable elements. That a natural system is complex does not mean that predictions are not possible for some spatial, temporal and size regimes. Long-term, and perhaps intermediate-term, predictions for large earthquakes appear to be possible for very active fault segments. Predicting large events more than one cycle into the future appears to be inherently difficult, if not impossible since much of the nonlinearity in the earthquake process occurs at or near the time of large events. Progre